Relationship to timeseries concepts in other domains Relationship to ISO19123 – Coverages

OGC WaterML 2.0 OGC 10-126r4 Copyright © 2012-2014 Open Geospatial Consortium 37 The ISO coverages model describes two approaches to representing coverages: a ‘domain-range’ representation where the domain and range are encoded separately, with a mapping function that allows looking up of the range value for a given domain value; and a ‘geometry-value’, or interleaved, approach whereby the geometry and value are coupled together – the coupling explicitly represents the mapping. GML 3.2.1 notes that the geometry-value approach “... is typically used during data collection where a set or properties relating to a single location are managed together, or update of a datastore where only a small number of features are manipulated at one time.” And the domain-range approach is ‘…more suitable for analysis, where spatio-temporal patterns and anomalies within a specific property are of interest.” Within hydrology this is often the case. For example, a grid showing the spatial distribution of rainfall is often generated from observations using interpolation techniques such as kriging. The surface may be generated using point observations from in-situ sensors. The point observations are often represented using a geometry-value structure with the generated surface being represented using the domain-range approach, with a spatial grid domain mapped to its range values representing total rainfall in the grid cell, for example. This provides a more efficient representation. WaterML2.0 defines a timeseries as a coverage whose domain consists of collection of ordered temporal elements and the spatial component relates to the feature of interest of the observation. For in-situ timeseries the spatial element will be fixed and need not be directly represented in the timeseries domain. The core coverage elements and the relationship to timeseries are shown in Figure 16 and Figure 18. Temporal axis Domain Parameter Space Range Figure 17 - Timeseries as a coverage Copyright © 2012-2014 Open Geospatial Consortium 38 Figure 18 - Timeseries as a coverage A timeseries may then be viewed in two ways from a coverage perspective: using the ‘domain- range’ view or the ‘geometry-value’ or interleaved view. Note that the term ‘geometry’ holds the domain object and is composed of varying spatial and temporal components e.g. time instants. The two types are show in Figure 21 and Figure 22 respectively. The geometry-value view is consistent with the most common structuring in the hydrology domain: time and values are coupled together and represent discrete observations at time instants. The use of the term geometry is based on the coverage viewpoint; time-value will be used in place for clarity.

9.12.3 Timeseries and point metadata

Associating metadata with timeseries as a whole and at each individual point is a common requirement in hydrological data. Data is annotated with various types of qualifying information such as quality assertions, affecting environmental conditions, description of processing and so on. These annotations are important when processing and analysing timeseries to ensure correct interpretation. At a generic level it is possible to associate any metadata with timeseries and timeseries points, this is shown with Annotation associations in Figure 21 and Figure 22. Using a soft-typed approach, this is simply a collection of named value pairs. Whilst flexible, this approach doesn’t «type» Coverage Core::CV_Coverage + domai Ã Ä Å tent : Ä Æ Ç Ä Å tent [1..] + rangeType :RecordType + commonPointRule :CV_CommonPointRule + evaluat È É DirectPosition Ê S È Ë Ì È Ã Í È Î CharacterStri Ã Ï Ð Ñ :Record + evaluateInvers È É Record Ñ :Set Î CV_DomainO Ò Ó ect Ð + fi Ã Ô É DirectPosition Ê Integer Ñ :S È Ë Ì È Ã Í È Î CV_GeometryValuePair Ð + list É Ñ :Set Î CV_GeometryValuePair Ð + select É GM_O Ò Ó ect Ê TM_Period Ñ :Set Î CV_GeometryValuePair Ð «type» Cov erage Core::CV_DomainObj ect «type» Geometry root:: GM_Object Õ root Ö × M Ø Ù r Ú Û Ú Ü Ú Ý e Temporal Obj ects:: TM_GeometricPrimitiv e «Type» WML_DomainObj ect constraints Õ temporal Ä lements shall be ordered in increasing tim È Ö Õ result domain = feature of interest + time series Ö «type» Cov erage Core:: CV_AttributeValues + values :Record +domai Ã Ä lement 1.. Domain +collection +spatial Ä lement 0.. SpatialComposition +temporal Ä lement 0.. TemporalComposition SpatialComposition +spatial Ä lement 1 TemporalComposition +temporal Ä lement 1 +r Þ Ã Ï È Ä lement 0.. Range +collection OGC WaterML 2.0 OGC 10-126r4 Copyright © 2012-2014 Open Geospatial Consortium 39 capture semantics specific to the metadata elements. WaterML2.0 thus defines two specialisations for timeseries and point-based metadata as shown in Figure 19 and Figure 20. Using the interleaved structuring, metadata is associated with a time-value pair explicitly, with the metadata directly associated with a pair. Using the domain-range structuring, metadata is associated to the timeseries through an AnnotationCoverage. This is a coverage that describes the temporally ranging metadata for the series. The domain of the coverage is the time that the annotation is valid; the range captures the values of the annotation e.g. a quality assertion. In ISO19123 the values of range are described using a Record, which is a generic set of typed values. Each annotation that is required would need an associate record type. For example, quality may be a string record type that allows for simple categorical representations of quality. The core elements of annotation for timeseries and timeseries points are shown in Figure 19 and Figure 20. These types are further extended for specific series types e.g. measurement series. Each of the metadata elements is described in the following sections.

9.12.3.1 Timeseries metadata

Figure 19 - Timeseries metadata The following sections define the available metadata properties for timeseries.

9.12.3.1.1 domai xtent

The domain extent is the temporal extent of the timeseries. The concept is inherited from the coverage model as shown in Figure 18. As the domain of the timeseries is temporal, the domainExtent is a time period defining the start and end of its temporal domain i.e. the start and end of the timeseries. Note that this often the same as the phenomenon time as specified in the OM_Observation; it is still useful here for timeseries that are described separately from an OM_Observation header.

9.12.3.1.2 quidistant time seri

baseTime and spacing Time series that are regularly spaced, such as those that are generated from automatic sensors, can be represented without specifying the individual time instant for each point. The spacing property of the time series is used to specify the time between points. This is then used as the spacing for each point encountered, starting from the time set by baseTime. See requirement reqxsd-timeseries-tvpequidistant-encoding in section 10.8 for details on how this is implemented in XML. If the spacing between the timeseries values differs then the time instants «DataType» TimeseriesMetadata + baseTime :TM_Instant [0..1] + spacing :TM_PeriodDuration [0..1] + domai tent :TM_Period [0..1] + parameter :NamedValue [0..]